AI summary

This is a local software benchmark. It compares MemO(1) against Mem0 using server Qdrant on the same workstation, same telemetry-derived query set, and the same local Qwen 2.5 7B answer model.

At 1000 history turns on exact recall, MemO(1) retrieval was 100.0% accurate at 0.24 ms p95. Mem0 retrieval was 0.2% accurate at 152.8548 ms p95.

At 1000 history turns on structured telemetry questions, MemO(1) retrieval was 100.0% accurate at 0.6548 ms p95. Mem0 retrieval was 10.2% top-1 accurate, 28.0% top-5 accurate, at 189.241 ms p95.

On answer generation for the same 1000-turn structured task, MemO(1) used 218.86 average Qwen prompt tokens. Mem0 used 696.4 average Qwen prompt tokens.

Measured caveat: this is structured routing vs embedding search, not a generic semantic-search benchmark. It does not claim FPGA timing.